Autonomous agents, Chinese models: the Fed, the OCC and the FDIC are questioning banks about their use of AI

The three US banking regulators are questioning institutions about their oversight of AI systems, whilst the Bank of England and the ECB are pushing for stronger disaster recovery plans.

According to six sources cited by Reuters, the Federal Reserve, the Office of the Comptroller of the Currency and the FDIC have been sending written and oral questions to American banks, large and small, for several weeks about their use of advanced AI. Supervisors want to know if they are using models developed in China, if autonomous agents are involved in their cyber defense, what limits are imposed on them, how their decisions are plotted and validated, and if the teams know how to cut them in the event of drift. At this stage, it is a census: no new requirements have been formulated.

Europe bets on recovery capacity

On this side of the Atlantic, the Bank of England and the ECB are urging institutions to improve their recovery arrangements, fearing that advances in AI could cause disruption in markets sooner than expected. British supervisors are encouraging the largest banks to study a so-called "bare metal" recovery, on an infrastructure completely separate from the main system, a costly solution. The ECB had set a deadline of the end of October in July to receive banks' plans against AI-boosted cyber threats.

This tightening comes after several recent incidents, including the hacking of South Korean banks attributed by CrowdStrike to a Chinese actor relying on AI agents.